When partial discharge (PD) measurement is used to monitor the power cables and analyze the insulation conditions, the collected signals often contains a variety of noises, in which white noise is the most common and most influential type. In order to suppress the influence of white noise, a white noise reduction method for PD signal based on Variational Mode Decomposition (VMD) is proposed. Firstly, the PD signal is decomposed by VMD, which obtains Intrinsic Mode Functions (IMFs) with low to high frequency. Then the kurtosis value of each IMF is calculated respectively. The corresponding IMFs are selected to be reconstructed. And the reconstructed signal is de-noised again by the wavelet unbiased estimation using adaptive threshold. Compared with different noise environments and de-noising methods, the results show that the de-noising effect about errors of mean square and quantity of waveform similarity coefficient based on VMD is better than that of wavelet standard soft threshold de-noising method or global hard threshold method. The simulation calculations and monitoring experiments show that this method can not only effectively remove the noise signal, but also preserve the original signal waveform completely. This work is supported by Science and Technology Support Xinjiang Project of Xinjiang Uygur Autonomous Region (No. 2018E02073), Henan Science and Technology Key Project (No. 182102310936), and Project of the Quality and Technological Supervision of Henan Province (No. 2018ZJ52).